Vishnu Sharma

Architected an intelligent AI agent classification and dynamic routing engine that analyzes trace outputs to categorize agent behavior, offloading low-complexity tasks to lightweight models or deterministic code paths to reduce LLM API calls, inference costs, and SLA latency. Established a standardized error taxonomy across AI evaluation services, enabling systematic failure classification and reducing root-cause analysis time across distributed agent workflows. Engineered an automated integration test suite using Vitest across multi-service evaluation boundaries, improving release confidence and eliminating regressions in production releases. Partnered directly with enterprise clients to design and deploy end-to-end LLM evaluation pipelines for production applications, aligning platform capabilities with client reliability standards. Collaborated cross-functionally with product and platform engineering teams to integrate AI evaluation tooling and real-time observability features into client production environments.

Integrated Gemini AI to auto-transcribe and analyze post-video KYC sessions, reducing average manual review time from 7 minutes to under 2 minutes per session leading to 75%+ reduction in operational effort and near-instant turnaround. Engineered a multi-level conditional profile allocation system, improving resource utilization across operations. Driving technical design and mentoring juniors on real-time architecture, and distributed system best practices. Driving cross-functional technical planning to scale IDfys KYC and onboarding platforms, optimizing reliability for high-throughput real-time workflows.

Designed a custom archival service reducing production DB load and improving GET API response time by 60%. Led development of video-based personal discussion and e-signature workflows, contributing to a 12% revenue boost and onboarding of major clients like PhonePe and Godrej. Boosted client conversion rates by 10% by integrating MeriPehchaans Pinless DigiLocker into verification flows. Integrated our Workflows with our own in-house Consent Gateway for regulatory compliance with the Data Protection Act. Technologies: Elixir, Phoenix, PostgreSQL, Redis, RabbitMQ, Docker, Kubernetes, GCP, React.

Created interactive ClickHouse dashboards via Apache Beam pipelines, improving data-driven decision-making. Redesigned customer-agent allocation logic, increasing assignment accuracy and improving support efficiency. Rolled out real-time alerting and monitoring via Grafana. Worked on instrumentations, driving better analytics and improved customer targeting. Technologies: Go, Apache Beam, ClickHouse, RabbitMQ, Grafana, Docker.